drivers/xen/events.c

   1 /*
   2  * Xen event channels
   3  *
   4  * Xen models interrupts with abstract event channels.  Because each
   5  * domain gets 1024 event channels, but NR_IRQ is not that large, we
   6  * must dynamically map irqs<->event channels.  The event channels
   7  * interface with the rest of the kernel by defining a xen interrupt
   8  * chip.  When an event is recieved, it is mapped to an irq and sent
   9  * through the normal interrupt processing path.
  10  *
  11  * There are four kinds of events which can be mapped to an event
  12  * channel:
  13  *
  14  * 1. Inter-domain notifications.  This includes all the virtual
  15  *    device events, since they're driven by front-ends in another domain
  16  *    (typically dom0).
  17  * 2. VIRQs, typically used for timers.  These are per-cpu events.
  18  * 3. IPIs.
  19  * 4. PIRQs - Hardware interrupts.
  20  *
  21  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  22  */
  23
  24 #include <linux/linkage.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/irq.h>
  27 #include <linux/module.h>
  28 #include <linux/string.h>
  29 #include <linux/bootmem.h>
  30 #include <linux/slab.h>
  31 #include <linux/irqnr.h>
  32 #include <linux/pci.h>
  33
  34 #include <asm/desc.h>
  35 #include <asm/ptrace.h>
  36 #include <asm/irq.h>
  37 #include <asm/idle.h>
  38 #include <asm/io_apic.h>
  39 #include <asm/sync_bitops.h>
  40 #include <asm/xen/pci.h>
  41 #include <asm/xen/hypercall.h>
  42 #include <asm/xen/hypervisor.h>
  43
  44 #include <xen/xen.h>
  45 #include <xen/hvm.h>
  46 #include <xen/xen-ops.h>
  47 #include <xen/events.h>
  48 #include <xen/interface/xen.h>
  49 #include <xen/interface/event_channel.h>
  50 #include <xen/interface/hvm/hvm_op.h>
  51 #include <xen/interface/hvm/params.h>
  52
  53 /*
  54  * This lock protects updates to the following mapping and reference-count
  55  * arrays. The lock does not need to be acquired to read the mapping tables.
  56  */
  57 static DEFINE_SPINLOCK(irq_mapping_update_lock);
  58
  59 /* IRQ <-> VIRQ mapping. */
  60 static DEFINE_PER_CPU(int [NR_VIRQS], virq_to_irq) = {[0 ... NR_VIRQS-1] = -1};
  61
  62 /* IRQ <-> IPI mapping */
  63 static DEFINE_PER_CPU(int [XEN_NR_IPIS], ipi_to_irq) = {[0 ... XEN_NR_IPIS-1] = -1};
  64
  65 /* Interrupt types. */
  66 enum xen_irq_type {
  67         IRQT_UNBOUND = 0,
  68         IRQT_PIRQ,
  69         IRQT_VIRQ,
  70         IRQT_IPI,
  71         IRQT_EVTCHN
  72 };
  73
  74 /*
  75  * Packed IRQ information:
  76  * type - enum xen_irq_type
  77  * event channel - irq->event channel mapping
  78  * cpu - cpu this event channel is bound to
  79  * index - type-specific information:
  80  *    PIRQ - vector, with MSB being "needs EIO", or physical IRQ of the HVM
  81  *           guest, or GSI (real passthrough IRQ) of the device.
  82  *    VIRQ - virq number
  83  *    IPI - IPI vector
  84  *    EVTCHN -
  85  */
  86 struct irq_info
  87 {
  88         enum xen_irq_type type; /* type */
  89         unsigned short evtchn;  /* event channel */
  90         unsigned short cpu;     /* cpu bound */
  91
  92         union {
  93                 unsigned short virq;
  94                 enum ipi_vector ipi;
  95                 struct {
  96                         unsigned short pirq;
  97                         unsigned short gsi;
  98                         unsigned char vector;
  99                         unsigned char flags;
 100                 } pirq;
 101         } u;
 102 };
 103 #define PIRQ_NEEDS_EOI  (1 << 0)
 104 #define PIRQ_SHAREABLE  (1 << 1)
 105
 106 static struct irq_info *irq_info;
 107 static int *pirq_to_irq;
 108
 109 static int *evtchn_to_irq;
 110 struct cpu_evtchn_s {
 111         unsigned long bits[NR_EVENT_CHANNELS/BITS_PER_LONG];
 112 };
 113
 114 static __initdata struct cpu_evtchn_s init_evtchn_mask = {
 115         .bits[0 ... (NR_EVENT_CHANNELS/BITS_PER_LONG)-1] = ~0ul,
 116 };
 117 static struct cpu_evtchn_s *cpu_evtchn_mask_p = &init_evtchn_mask;
 118
 119 static inline unsigned long *cpu_evtchn_mask(int cpu)
 120 {
 121         return cpu_evtchn_mask_p[cpu].bits;
 122 }
 123
 124 /* Xen will never allocate port zero for any purpose. */
 125 #define VALID_EVTCHN(chn)       ((chn) != 0)
 126
 127 static struct irq_chip xen_dynamic_chip;
 128 static struct irq_chip xen_percpu_chip;
 129 static struct irq_chip xen_pirq_chip;
 130
 131 /* Constructor for packed IRQ information. */
 132 static struct irq_info mk_unbound_info(void)
 133 {
 134         return (struct irq_info) { .type = IRQT_UNBOUND };
 135 }
 136
 137 static struct irq_info mk_evtchn_info(unsigned short evtchn)
 138 {
 139         return (struct irq_info) { .type = IRQT_EVTCHN, .evtchn = evtchn,
 140                         .cpu = 0 };
 141 }
 142
 143 static struct irq_info mk_ipi_info(unsigned short evtchn, enum ipi_vector ipi)
 144 {
 145         return (struct irq_info) { .type = IRQT_IPI, .evtchn = evtchn,
 146                         .cpu = 0, .u.ipi = ipi };
 147 }
 148
 149 static struct irq_info mk_virq_info(unsigned short evtchn, unsigned short virq)
 150 {
 151         return (struct irq_info) { .type = IRQT_VIRQ, .evtchn = evtchn,
 152                         .cpu = 0, .u.virq = virq };
 153 }
 154
 155 static struct irq_info mk_pirq_info(unsigned short evtchn, unsigned short pirq,
 156                                     unsigned short gsi, unsigned short vector)
 157 {
 158         return (struct irq_info) { .type = IRQT_PIRQ, .evtchn = evtchn,
 159                         .cpu = 0,
 160                         .u.pirq = { .pirq = pirq, .gsi = gsi, .vector = vector } };
 161 }
 162
 163 /*
 164  * Accessors for packed IRQ information.
 165  */
 166 static struct irq_info *info_for_irq(unsigned irq)
 167 {
 168         return &irq_info[irq];
 169 }
 170
 171 static unsigned int evtchn_from_irq(unsigned irq)
 172 {
 173         if (unlikely(WARN(irq < 0 || irq >= nr_irqs, "Invalid irq %d!\n", irq)))
 174                 return 0;
 175
 176         return info_for_irq(irq)->evtchn;
 177 }
 178
 179 unsigned irq_from_evtchn(unsigned int evtchn)
 180 {
 181         return evtchn_to_irq[evtchn];
 182 }
 183 EXPORT_SYMBOL_GPL(irq_from_evtchn);
 184
 185 static enum ipi_vector ipi_from_irq(unsigned irq)
 186 {
 187         struct irq_info *info = info_for_irq(irq);
 188
 189         BUG_ON(info == NULL);
 190         BUG_ON(info->type != IRQT_IPI);
 191
 192         return info->u.ipi;
 193 }
 194
 195 static unsigned virq_from_irq(unsigned irq)
 196 {
 197         struct irq_info *info = info_for_irq(irq);
 198
 199         BUG_ON(info == NULL);
 200         BUG_ON(info->type != IRQT_VIRQ);
 201
 202         return info->u.virq;
 203 }
 204
 205 static unsigned pirq_from_irq(unsigned irq)
 206 {
 207         struct irq_info *info = info_for_irq(irq);
 208
 209         BUG_ON(info == NULL);
 210         BUG_ON(info->type != IRQT_PIRQ);
 211
 212         return info->u.pirq.pirq;
 213 }
 214
 215 static unsigned gsi_from_irq(unsigned irq)
 216 {
 217         struct irq_info *info = info_for_irq(irq);
 218
 219         BUG_ON(info == NULL);
 220         BUG_ON(info->type != IRQT_PIRQ);
 221
 222         return info->u.pirq.gsi;
 223 }
 224
 225 static unsigned vector_from_irq(unsigned irq)
 226 {
 227         struct irq_info *info = info_for_irq(irq);
 228
 229         BUG_ON(info == NULL);
 230         BUG_ON(info->type != IRQT_PIRQ);
 231
 232         return info->u.pirq.vector;
 233 }
 234
 235 static enum xen_irq_type type_from_irq(unsigned irq)
 236 {
 237         return info_for_irq(irq)->type;
 238 }
 239
 240 static unsigned cpu_from_irq(unsigned irq)
 241 {
 242         return info_for_irq(irq)->cpu;
 243 }
 244
 245 static unsigned int cpu_from_evtchn(unsigned int evtchn)
 246 {
 247         int irq = evtchn_to_irq[evtchn];
 248         unsigned ret = 0;
 249
 250         if (irq != -1)
 251                 ret = cpu_from_irq(irq);
 252
 253         return ret;
 254 }
 255
 256 static bool pirq_needs_eoi(unsigned irq)
 257 {
 258         struct irq_info *info = info_for_irq(irq);
 259
 260         BUG_ON(info->type != IRQT_PIRQ);
 261
 262         return info->u.pirq.flags & PIRQ_NEEDS_EOI;
 263 }
 264
 265 static inline unsigned long active_evtchns(unsigned int cpu,
 266                                            struct shared_info *sh,
 267                                            unsigned int idx)
 268 {
 269         return (sh->evtchn_pending[idx] &
 270                 cpu_evtchn_mask(cpu)[idx] &
 271                 ~sh->evtchn_mask[idx]);
 272 }
 273
 274 static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu)
 275 {
 276         int irq = evtchn_to_irq[chn];
 277
 278         BUG_ON(irq == -1);
 279 #ifdef CONFIG_SMP
 280         cpumask_copy(irq_to_desc(irq)->irq_data.affinity, cpumask_of(cpu));
 281 #endif
 282
 283         clear_bit(chn, cpu_evtchn_mask(cpu_from_irq(irq)));
 284         set_bit(chn, cpu_evtchn_mask(cpu));
 285
 286         irq_info[irq].cpu = cpu;
 287 }
 288
 289 static void init_evtchn_cpu_bindings(void)
 290 {
 291         int i;
 292 #ifdef CONFIG_SMP
 293         struct irq_desc *desc;
 294
 295         /* By default all event channels notify CPU#0. */
 296         for_each_irq_desc(i, desc) {
 297                 cpumask_copy(desc->irq_data.affinity, cpumask_of(0));
 298         }
 299 #endif
 300
 301         for_each_possible_cpu(i)
 302                 memset(cpu_evtchn_mask(i),
 303                        (i == 0) ? ~0 : 0, sizeof(struct cpu_evtchn_s));
 304
 305 }
 306
 307 static inline void clear_evtchn(int port)
 308 {
 309         struct shared_info *s = HYPERVISOR_shared_info;
 310         sync_clear_bit(port, &s->evtchn_pending[0]);
 311 }
 312
 313 static inline void set_evtchn(int port)
 314 {
 315         struct shared_info *s = HYPERVISOR_shared_info;
 316         sync_set_bit(port, &s->evtchn_pending[0]);
 317 }
 318
 319 static inline int test_evtchn(int port)
 320 {
 321         struct shared_info *s = HYPERVISOR_shared_info;
 322         return sync_test_bit(port, &s->evtchn_pending[0]);
 323 }
 324
 325
 326 /**
 327  * notify_remote_via_irq - send event to remote end of event channel via irq
 328  * @irq: irq of event channel to send event to
 329  *
 330  * Unlike notify_remote_via_evtchn(), this is safe to use across
 331  * save/restore. Notifications on a broken connection are silently
 332  * dropped.
 333  */
 334 void notify_remote_via_irq(int irq)
 335 {
 336         int evtchn = evtchn_from_irq(irq);
 337
 338         if (VALID_EVTCHN(evtchn))
 339                 notify_remote_via_evtchn(evtchn);
 340 }
 341 EXPORT_SYMBOL_GPL(notify_remote_via_irq);
 342
 343 static void mask_evtchn(int port)
 344 {
 345         struct shared_info *s = HYPERVISOR_shared_info;
 346         sync_set_bit(port, &s->evtchn_mask[0]);
 347 }
 348
 349 static void unmask_evtchn(int port)
 350 {
 351         struct shared_info *s = HYPERVISOR_shared_info;
 352         unsigned int cpu = get_cpu();
 353
 354         BUG_ON(!irqs_disabled());
 355
 356         /* Slow path (hypercall) if this is a non-local port. */
 357         if (unlikely(cpu != cpu_from_evtchn(port))) {
 358                 struct evtchn_unmask unmask = { .port = port };
 359                 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
 360         } else {
 361                 struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
 362
 363                 sync_clear_bit(port, &s->evtchn_mask[0]);
 364
 365                 /*
 366                  * The following is basically the equivalent of
 367                  * 'hw_resend_irq'. Just like a real IO-APIC we 'lose
 368                  * the interrupt edge' if the channel is masked.
 369                  */
 370                 if (sync_test_bit(port, &s->evtchn_pending[0]) &&
 371                     !sync_test_and_set_bit(port / BITS_PER_LONG,
 372                                            &vcpu_info->evtchn_pending_sel))
 373                         vcpu_info->evtchn_upcall_pending = 1;
 374         }
 375
 376         put_cpu();
 377 }
 378
 379 static int xen_allocate_irq_dynamic(void)
 380 {
 381         int first = 0;
 382         int irq;
 383
 384 #ifdef CONFIG_X86_IO_APIC
 385         /*
 386          * For an HVM guest or domain 0 which see "real" (emulated or
 387          * actual repectively) GSIs we allocate dynamic IRQs
 388          * e.g. those corresponding to event channels or MSIs
 389          * etc. from the range above those "real" GSIs to avoid
 390          * collisions.
 391          */
 392         if (xen_initial_domain() || xen_hvm_domain())
 393                 first = get_nr_irqs_gsi();
 394 #endif
 395
 396 retry:
 397         irq = irq_alloc_desc_from(first, -1);
 398
 399         if (irq == -ENOMEM && first > NR_IRQS_LEGACY) {
 400                 printk(KERN_ERR "Out of dynamic IRQ space and eating into GSI space. You should increase nr_irqs\n");
 401                 first = max(NR_IRQS_LEGACY, first - NR_IRQS_LEGACY);
 402                 goto retry;
 403         }
 404
 405         if (irq < 0)
 406                 panic("No available IRQ to bind to: increase nr_irqs!\n");
 407
 408         return irq;
 409 }
 410
 411 static int xen_allocate_irq_gsi(unsigned gsi)
 412 {
 413         int irq;
 414
 415         /*
 416          * A PV guest has no concept of a GSI (since it has no ACPI
 417          * nor access to/knowledge of the physical APICs). Therefore
 418          * all IRQs are dynamically allocated from the entire IRQ
 419          * space.
 420          */
 421         if (xen_pv_domain() && !xen_initial_domain())
 422                 return xen_allocate_irq_dynamic();
 423
 424         /* Legacy IRQ descriptors are already allocated by the arch. */
 425         if (gsi < NR_IRQS_LEGACY)
 426                 return gsi;
 427
 428         irq = irq_alloc_desc_at(gsi, -1);
 429         if (irq < 0)
 430                 panic("Unable to allocate to IRQ%d (%d)\n", gsi, irq);
 431
 432         return irq;
 433 }
 434
 435 static void xen_free_irq(unsigned irq)
 436 {
 437         /* Legacy IRQ descriptors are managed by the arch. */
 438         if (irq < NR_IRQS_LEGACY)
 439                 return;
 440
 441         irq_free_desc(irq);
 442 }
 443
 444 static void pirq_unmask_notify(int irq)
 445 {
 446         struct physdev_eoi eoi = { .irq = pirq_from_irq(irq) };
 447
 448         if (unlikely(pirq_needs_eoi(irq))) {
 449                 int rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
 450                 WARN_ON(rc);
 451         }
 452 }
 453
 454 static void pirq_query_unmask(int irq)
 455 {
 456         struct physdev_irq_status_query irq_status;
 457         struct irq_info *info = info_for_irq(irq);
 458
 459         BUG_ON(info->type != IRQT_PIRQ);
 460
 461         irq_status.irq = pirq_from_irq(irq);
 462         if (HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query, &irq_status))
 463                 irq_status.flags = 0;
 464
 465         info->u.pirq.flags &= ~PIRQ_NEEDS_EOI;
 466         if (irq_status.flags & XENIRQSTAT_needs_eoi)
 467                 info->u.pirq.flags |= PIRQ_NEEDS_EOI;
 468 }
 469
 470 static bool probing_irq(int irq)
 471 {
 472         struct irq_desc *desc = irq_to_desc(irq);
 473
 474         return desc && desc->action == NULL;
 475 }
 476
 477 static unsigned int __startup_pirq(unsigned int irq)
 478 {
 479         struct evtchn_bind_pirq bind_pirq;
 480         struct irq_info *info = info_for_irq(irq);
 481         int evtchn = evtchn_from_irq(irq);
 482         int rc;
 483
 484         BUG_ON(info->type != IRQT_PIRQ);
 485
 486         if (VALID_EVTCHN(evtchn))
 487                 goto out;
 488
 489         bind_pirq.pirq = pirq_from_irq(irq);
 490         /* NB. We are happy to share unless we are probing. */
 491         bind_pirq.flags = info->u.pirq.flags & PIRQ_SHAREABLE ?
 492                                         BIND_PIRQ__WILL_SHARE : 0;
 493         rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
 494         if (rc != 0) {
 495                 if (!probing_irq(irq))
 496                         printk(KERN_INFO "Failed to obtain physical IRQ %d\n",
 497                                irq);
 498                 return 0;
 499         }
 500         evtchn = bind_pirq.port;
 501
 502         pirq_query_unmask(irq);
 503
 504         evtchn_to_irq[evtchn] = irq;
 505         bind_evtchn_to_cpu(evtchn, 0);
 506         info->evtchn = evtchn;
 507
 508 out:
 509         unmask_evtchn(evtchn);
 510         pirq_unmask_notify(irq);
 511
 512         return 0;
 513 }
 514
 515 static unsigned int startup_pirq(struct irq_data *data)
 516 {
 517         return __startup_pirq(data->irq);
 518 }
 519
 520 static void shutdown_pirq(struct irq_data *data)
 521 {
 522         struct evtchn_close close;
 523         unsigned int irq = data->irq;
 524         struct irq_info *info = info_for_irq(irq);
 525         int evtchn = evtchn_from_irq(irq);
 526
 527         BUG_ON(info->type != IRQT_PIRQ);
 528
 529         if (!VALID_EVTCHN(evtchn))
 530                 return;
 531
 532         mask_evtchn(evtchn);
 533
 534         close.port = evtchn;
 535         if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 536                 BUG();
 537
 538         bind_evtchn_to_cpu(evtchn, 0);
 539         evtchn_to_irq[evtchn] = -1;
 540         info->evtchn = 0;
 541 }
 542
 543 static void enable_pirq(struct irq_data *data)
 544 {
 545         startup_pirq(data);
 546 }
 547
 548 static void disable_pirq(struct irq_data *data)
 549 {
 550 }
 551
 552 static void ack_pirq(struct irq_data *data)
 553 {
 554         int evtchn = evtchn_from_irq(data->irq);
 555
 556         move_native_irq(data->irq);
 557
 558         if (VALID_EVTCHN(evtchn)) {
 559                 mask_evtchn(evtchn);
 560                 clear_evtchn(evtchn);
 561         }
 562 }
 563
 564 static int find_irq_by_gsi(unsigned gsi)
 565 {
 566         int irq;
 567
 568         for (irq = 0; irq < nr_irqs; irq++) {
 569                 struct irq_info *info = info_for_irq(irq);
 570
 571                 if (info == NULL || info->type != IRQT_PIRQ)
 572                         continue;
 573
 574                 if (gsi_from_irq(irq) == gsi)
 575                         return irq;
 576         }
 577
 578         return -1;
 579 }
 580
 581 int xen_allocate_pirq(unsigned gsi, int shareable, char *name)
 582 {
 583         return xen_map_pirq_gsi(gsi, gsi, shareable, name);
 584 }
 585
 586 /* xen_map_pirq_gsi might allocate irqs from the top down, as a
 587  * consequence don't assume that the irq number returned has a low value
 588  * or can be used as a pirq number unless you know otherwise.
 589  *
 590  * One notable exception is when xen_map_pirq_gsi is called passing an
 591  * hardware gsi as argument, in that case the irq number returned
 592  * matches the gsi number passed as second argument.
 593  *
 594  * Note: We don't assign an event channel until the irq actually started
 595  * up.  Return an existing irq if we've already got one for the gsi.
 596  */
 597 int xen_map_pirq_gsi(unsigned pirq, unsigned gsi, int shareable, char *name)
 598 {
 599         int irq = 0;
 600         struct physdev_irq irq_op;
 601
 602         spin_lock(&irq_mapping_update_lock);
 603
 604         if ((pirq > nr_irqs) || (gsi > nr_irqs)) {
 605                 printk(KERN_WARNING "xen_map_pirq_gsi: %s %s is incorrect!\n",
 606                         pirq > nr_irqs ? "pirq" :"",
 607                         gsi > nr_irqs ? "gsi" : "");
 608                 goto out;
 609         }
 610
 611         irq = find_irq_by_gsi(gsi);
 612         if (irq != -1) {
 613                 printk(KERN_INFO "xen_map_pirq_gsi: returning irq %d for gsi %u\n",
 614                        irq, gsi);
 615                 goto out;       /* XXX need refcount? */
 616         }
 617
 618         irq = xen_allocate_irq_gsi(gsi);
 619
 620         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 621                                       handle_level_irq, name);
 622
 623         irq_op.irq = irq;
 624         irq_op.vector = 0;
 625
 626         /* Only the privileged domain can do this. For non-priv, the pcifront
 627          * driver provides a PCI bus that does the call to do exactly
 628          * this in the priv domain. */
 629         if (xen_initial_domain() &&
 630             HYPERVISOR_physdev_op(PHYSDEVOP_alloc_irq_vector, &irq_op)) {
 631                 xen_free_irq(irq);
 632                 irq = -ENOSPC;
 633                 goto out;
 634         }
 635
 636         irq_info[irq] = mk_pirq_info(0, pirq, gsi, irq_op.vector);
 637         irq_info[irq].u.pirq.flags |= shareable ? PIRQ_SHAREABLE : 0;
 638         pirq_to_irq[pirq] = irq;
 639
 640 out:
 641         spin_unlock(&irq_mapping_update_lock);
 642
 643         return irq;
 644 }
 645
 646 #ifdef CONFIG_PCI_MSI
 647 #include <linux/msi.h>
 648 #include "../pci/msi.h"
 649
 650 static int find_unbound_pirq(int type)
 651 {
 652         int rc, i;
 653         struct physdev_get_free_pirq op_get_free_pirq;
 654         op_get_free_pirq.type = type;
 655
 656         rc = HYPERVISOR_physdev_op(PHYSDEVOP_get_free_pirq, &op_get_free_pirq);
 657         if (!rc)
 658                 return op_get_free_pirq.pirq;
 659
 660         for (i = 0; i < nr_irqs; i++) {
 661                 if (pirq_to_irq[i] < 0)
 662                         return i;
 663         }
 664         return -1;
 665 }
 666
 667 void xen_allocate_pirq_msi(char *name, int *irq, int *pirq, int alloc)
 668 {
 669         spin_lock(&irq_mapping_update_lock);
 670
 671         if (alloc & XEN_ALLOC_IRQ) {
 672                 *irq = xen_allocate_irq_dynamic();
 673                 if (*irq == -1)
 674                         goto out;
 675         }
 676
 677         if (alloc & XEN_ALLOC_PIRQ) {
 678                 *pirq = find_unbound_pirq(MAP_PIRQ_TYPE_MSI);
 679                 if (*pirq == -1)
 680                         goto out;
 681         }
 682
 683         set_irq_chip_and_handler_name(*irq, &xen_pirq_chip,
 684                                       handle_level_irq, name);
 685
 686         irq_info[*irq] = mk_pirq_info(0, *pirq, 0, 0);
 687         pirq_to_irq[*pirq] = *irq;
 688
 689 out:
 690         spin_unlock(&irq_mapping_update_lock);
 691 }
 692
 693 int xen_create_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int type)
 694 {
 695         int irq = -1;
 696         struct physdev_map_pirq map_irq;
 697         int rc;
 698         int pos;
 699         u32 table_offset, bir;
 700
 701         memset(&map_irq, 0, sizeof(map_irq));
 702         map_irq.domid = DOMID_SELF;
 703         map_irq.type = MAP_PIRQ_TYPE_MSI;
 704         map_irq.index = -1;
 705         map_irq.pirq = -1;
 706         map_irq.bus = dev->bus->number;
 707         map_irq.devfn = dev->devfn;
 708
 709         if (type == PCI_CAP_ID_MSIX) {
 710                 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 711
 712                 pci_read_config_dword(dev, msix_table_offset_reg(pos),
 713                                         &table_offset);
 714                 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 715
 716                 map_irq.table_base = pci_resource_start(dev, bir);
 717                 map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 718         }
 719
 720         spin_lock(&irq_mapping_update_lock);
 721
 722         irq = xen_allocate_irq_dynamic();
 723
 724         if (irq == -1)
 725                 goto out;
 726
 727         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
 728         if (rc) {
 729                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
 730
 731                 xen_free_irq(irq);
 732
 733                 irq = -1;
 734                 goto out;
 735         }
 736         irq_info[irq] = mk_pirq_info(0, map_irq.pirq, 0, map_irq.index);
 737
 738         set_irq_chip_and_handler_name(irq, &xen_pirq_chip,
 739                         handle_level_irq,
 740                         (type == PCI_CAP_ID_MSIX) ? "msi-x":"msi");
 741
 742 out:
 743         spin_unlock(&irq_mapping_update_lock);
 744         return irq;
 745 }
 746 #endif
 747
 748 int xen_destroy_irq(int irq)
 749 {
 750         struct irq_desc *desc;
 751         struct physdev_unmap_pirq unmap_irq;
 752         struct irq_info *info = info_for_irq(irq);
 753         int rc = -ENOENT;
 754
 755         spin_lock(&irq_mapping_update_lock);
 756
 757         desc = irq_to_desc(irq);
 758         if (!desc)
 759                 goto out;
 760
 761         if (xen_initial_domain()) {
 762                 unmap_irq.pirq = info->u.pirq.pirq;
 763                 unmap_irq.domid = DOMID_SELF;
 764                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap_irq);
 765                 if (rc) {
 766                         printk(KERN_WARNING "unmap irq failed %d\n", rc);
 767                         goto out;
 768                 }
 769         }
 770         pirq_to_irq[info->u.pirq.pirq] = -1;
 771
 772         irq_info[irq] = mk_unbound_info();
 773
 774         xen_free_irq(irq);
 775
 776 out:
 777         spin_unlock(&irq_mapping_update_lock);
 778         return rc;
 779 }
 780
 781 int xen_vector_from_irq(unsigned irq)
 782 {
 783         return vector_from_irq(irq);
 784 }
 785
 786 int xen_gsi_from_irq(unsigned irq)
 787 {
 788         return gsi_from_irq(irq);
 789 }
 790
 791 int xen_irq_from_pirq(unsigned pirq)
 792 {
 793         return pirq_to_irq[pirq];
 794 }
 795
 796 int bind_evtchn_to_irq(unsigned int evtchn)
 797 {
 798         int irq;
 799
 800         spin_lock(&irq_mapping_update_lock);
 801
 802         irq = evtchn_to_irq[evtchn];
 803
 804         if (irq == -1) {
 805                 irq = xen_allocate_irq_dynamic();
 806
 807                 set_irq_chip_and_handler_name(irq, &xen_dynamic_chip,
 808                                               handle_fasteoi_irq, "event");
 809
 810                 evtchn_to_irq[evtchn] = irq;
 811                 irq_info[irq] = mk_evtchn_info(evtchn);
 812         }
 813
 814         spin_unlock(&irq_mapping_update_lock);
 815
 816         return irq;
 817 }
 818 EXPORT_SYMBOL_GPL(bind_evtchn_to_irq);
 819
 820 static int bind_ipi_to_irq(unsigned int ipi, unsigned int cpu)
 821 {
 822         struct evtchn_bind_ipi bind_ipi;
 823         int evtchn, irq;
 824
 825         spin_lock(&irq_mapping_update_lock);
 826
 827         irq = per_cpu(ipi_to_irq, cpu)[ipi];
 828
 829         if (irq == -1) {
 830                 irq = xen_allocate_irq_dynamic();
 831                 if (irq < 0)
 832                         goto out;
 833
 834                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 835                                               handle_percpu_irq, "ipi");
 836
 837                 bind_ipi.vcpu = cpu;
 838                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 839                                                 &bind_ipi) != 0)
 840                         BUG();
 841                 evtchn = bind_ipi.port;
 842
 843                 evtchn_to_irq[evtchn] = irq;
 844                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
 845                 per_cpu(ipi_to_irq, cpu)[ipi] = irq;
 846
 847                 bind_evtchn_to_cpu(evtchn, cpu);
 848         }
 849
 850  out:
 851         spin_unlock(&irq_mapping_update_lock);
 852         return irq;
 853 }
 854
 855
 856 int bind_virq_to_irq(unsigned int virq, unsigned int cpu)
 857 {
 858         struct evtchn_bind_virq bind_virq;
 859         int evtchn, irq;
 860
 861         spin_lock(&irq_mapping_update_lock);
 862
 863         irq = per_cpu(virq_to_irq, cpu)[virq];
 864
 865         if (irq == -1) {
 866                 irq = xen_allocate_irq_dynamic();
 867
 868                 set_irq_chip_and_handler_name(irq, &xen_percpu_chip,
 869                                               handle_percpu_irq, "virq");
 870
 871                 bind_virq.virq = virq;
 872                 bind_virq.vcpu = cpu;
 873                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 874                                                 &bind_virq) != 0)
 875                         BUG();
 876                 evtchn = bind_virq.port;
 877
 878                 evtchn_to_irq[evtchn] = irq;
 879                 irq_info[irq] = mk_virq_info(evtchn, virq);
 880
 881                 per_cpu(virq_to_irq, cpu)[virq] = irq;
 882
 883                 bind_evtchn_to_cpu(evtchn, cpu);
 884         }
 885
 886         spin_unlock(&irq_mapping_update_lock);
 887
 888         return irq;
 889 }
 890
 891 static void unbind_from_irq(unsigned int irq)
 892 {
 893         struct evtchn_close close;
 894         int evtchn = evtchn_from_irq(irq);
 895
 896         spin_lock(&irq_mapping_update_lock);
 897
 898         if (VALID_EVTCHN(evtchn)) {
 899                 close.port = evtchn;
 900                 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
 901                         BUG();
 902
 903                 switch (type_from_irq(irq)) {
 904                 case IRQT_VIRQ:
 905                         per_cpu(virq_to_irq, cpu_from_evtchn(evtchn))
 906                                 [virq_from_irq(irq)] = -1;
 907                         break;
 908                 case IRQT_IPI:
 909                         per_cpu(ipi_to_irq, cpu_from_evtchn(evtchn))
 910                                 [ipi_from_irq(irq)] = -1;
 911                         break;
 912                 default:
 913                         break;
 914                 }
 915
 916                 /* Closed ports are implicitly re-bound to VCPU0. */
 917                 bind_evtchn_to_cpu(evtchn, 0);
 918
 919                 evtchn_to_irq[evtchn] = -1;
 920         }
 921
 922         if (irq_info[irq].type != IRQT_UNBOUND) {
 923                 irq_info[irq] = mk_unbound_info();
 924
 925                 xen_free_irq(irq);
 926         }
 927
 928         spin_unlock(&irq_mapping_update_lock);
 929 }
 930
 931 int bind_evtchn_to_irqhandler(unsigned int evtchn,
 932                               irq_handler_t handler,
 933                               unsigned long irqflags,
 934                               const char *devname, void *dev_id)
 935 {
 936         unsigned int irq;
 937         int retval;
 938
 939         irq = bind_evtchn_to_irq(evtchn);
 940         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 941         if (retval != 0) {
 942                 unbind_from_irq(irq);
 943                 return retval;
 944         }
 945
 946         return irq;
 947 }
 948 EXPORT_SYMBOL_GPL(bind_evtchn_to_irqhandler);
 949
 950 int bind_virq_to_irqhandler(unsigned int virq, unsigned int cpu,
 951                             irq_handler_t handler,
 952                             unsigned long irqflags, const char *devname, void *dev_id)
 953 {
 954         unsigned int irq;
 955         int retval;
 956
 957         irq = bind_virq_to_irq(virq, cpu);
 958         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 959         if (retval != 0) {
 960                 unbind_from_irq(irq);
 961                 return retval;
 962         }
 963
 964         return irq;
 965 }
 966 EXPORT_SYMBOL_GPL(bind_virq_to_irqhandler);
 967
 968 int bind_ipi_to_irqhandler(enum ipi_vector ipi,
 969                            unsigned int cpu,
 970                            irq_handler_t handler,
 971                            unsigned long irqflags,
 972                            const char *devname,
 973                            void *dev_id)
 974 {
 975         int irq, retval;
 976
 977         irq = bind_ipi_to_irq(ipi, cpu);
 978         if (irq < 0)
 979                 return irq;
 980
 981         irqflags |= IRQF_NO_SUSPEND | IRQF_FORCE_RESUME;
 982         retval = request_irq(irq, handler, irqflags, devname, dev_id);
 983         if (retval != 0) {
 984                 unbind_from_irq(irq);
 985                 return retval;
 986         }
 987
 988         return irq;
 989 }
 990
 991 void unbind_from_irqhandler(unsigned int irq, void *dev_id)
 992 {
 993         free_irq(irq, dev_id);
 994         unbind_from_irq(irq);
 995 }
 996 EXPORT_SYMBOL_GPL(unbind_from_irqhandler);
 997
 998 void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector)
 999 {
1000         int irq = per_cpu(ipi_to_irq, cpu)[vector];
1001         BUG_ON(irq < 0);
1002         notify_remote_via_irq(irq);
1003 }
1004
1005 irqreturn_t xen_debug_interrupt(int irq, void *dev_id)
1006 {
1007         struct shared_info *sh = HYPERVISOR_shared_info;
1008         int cpu = smp_processor_id();
1009         unsigned long *cpu_evtchn = cpu_evtchn_mask(cpu);
1010         int i;
1011         unsigned long flags;
1012         static DEFINE_SPINLOCK(debug_lock);
1013         struct vcpu_info *v;
1014
1015         spin_lock_irqsave(&debug_lock, flags);
1016
1017         printk("\nvcpu %d\n  ", cpu);
1018
1019         for_each_online_cpu(i) {
1020                 int pending;
1021                 v = per_cpu(xen_vcpu, i);
1022                 pending = (get_irq_regs() && i == cpu)
1023                         ? xen_irqs_disabled(get_irq_regs())
1024                         : v->evtchn_upcall_mask;
1025                 printk("%d: masked=%d pending=%d event_sel %0*lx\n  ", i,
1026                        pending, v->evtchn_upcall_pending,
1027                        (int)(sizeof(v->evtchn_pending_sel)*2),
1028                        v->evtchn_pending_sel);
1029         }
1030         v = per_cpu(xen_vcpu, cpu);
1031
1032         printk("\npending:\n   ");
1033         for (i = ARRAY_SIZE(sh->evtchn_pending)-1; i >= 0; i--)
1034                 printk("%0*lx%s", (int)sizeof(sh->evtchn_pending[0])*2,
1035                        sh->evtchn_pending[i],
1036                        i % 8 == 0 ? "\n   " : " ");
1037         printk("\nglobal mask:\n   ");
1038         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1039                 printk("%0*lx%s",
1040                        (int)(sizeof(sh->evtchn_mask[0])*2),
1041                        sh->evtchn_mask[i],
1042                        i % 8 == 0 ? "\n   " : " ");
1043
1044         printk("\nglobally unmasked:\n   ");
1045         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--)
1046                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1047                        sh->evtchn_pending[i] & ~sh->evtchn_mask[i],
1048                        i % 8 == 0 ? "\n   " : " ");
1049
1050         printk("\nlocal cpu%d mask:\n   ", cpu);
1051         for (i = (NR_EVENT_CHANNELS/BITS_PER_LONG)-1; i >= 0; i--)
1052                 printk("%0*lx%s", (int)(sizeof(cpu_evtchn[0])*2),
1053                        cpu_evtchn[i],
1054                        i % 8 == 0 ? "\n   " : " ");
1055
1056         printk("\nlocally unmasked:\n   ");
1057         for (i = ARRAY_SIZE(sh->evtchn_mask)-1; i >= 0; i--) {
1058                 unsigned long pending = sh->evtchn_pending[i]
1059                         & ~sh->evtchn_mask[i]
1060                         & cpu_evtchn[i];
1061                 printk("%0*lx%s", (int)(sizeof(sh->evtchn_mask[0])*2),
1062                        pending, i % 8 == 0 ? "\n   " : " ");
1063         }
1064
1065         printk("\npending list:\n");
1066         for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1067                 if (sync_test_bit(i, sh->evtchn_pending)) {
1068                         int word_idx = i / BITS_PER_LONG;
1069                         printk("  %d: event %d -> irq %d%s%s%s\n",
1070                                cpu_from_evtchn(i), i,
1071                                evtchn_to_irq[i],
1072                                sync_test_bit(word_idx, &v->evtchn_pending_sel)
1073                                              ? "" : " l2-clear",
1074                                !sync_test_bit(i, sh->evtchn_mask)
1075                                              ? "" : " globally-masked",
1076                                sync_test_bit(i, cpu_evtchn)
1077                                              ? "" : " locally-masked");
1078                 }
1079         }
1080
1081         spin_unlock_irqrestore(&debug_lock, flags);
1082
1083         return IRQ_HANDLED;
1084 }
1085
1086 static DEFINE_PER_CPU(unsigned, xed_nesting_count);
1087
1088 /*
1089  * Search the CPUs pending events bitmasks.  For each one found, map
1090  * the event number to an irq, and feed it into do_IRQ() for
1091  * handling.
1092  *
1093  * Xen uses a two-level bitmap to speed searching.  The first level is
1094  * a bitset of words which contain pending event bits.  The second
1095  * level is a bitset of pending events themselves.
1096  */
1097 static void __xen_evtchn_do_upcall(void)
1098 {
1099         int cpu = get_cpu();
1100         struct shared_info *s = HYPERVISOR_shared_info;
1101         struct vcpu_info *vcpu_info = __this_cpu_read(xen_vcpu);
1102         unsigned count;
1103
1104         do {
1105                 unsigned long pending_words;
1106
1107                 vcpu_info->evtchn_upcall_pending = 0;
1108
1109                 if (__this_cpu_inc_return(xed_nesting_count) - 1)
1110                         goto out;
1111
1112 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
1113                 /* Clear master flag /before/ clearing selector flag. */
1114                 wmb();
1115 #endif
1116                 pending_words = xchg(&vcpu_info->evtchn_pending_sel, 0);
1117                 while (pending_words != 0) {
1118                         unsigned long pending_bits;
1119                         int word_idx = __ffs(pending_words);
1120                         pending_words &= ~(1UL << word_idx);
1121
1122                         while ((pending_bits = active_evtchns(cpu, s, word_idx)) != 0) {
1123                                 int bit_idx = __ffs(pending_bits);
1124                                 int port = (word_idx * BITS_PER_LONG) + bit_idx;
1125                                 int irq = evtchn_to_irq[port];
1126                                 struct irq_desc *desc;
1127
1128                                 mask_evtchn(port);
1129                                 clear_evtchn(port);
1130
1131                                 if (irq != -1) {
1132                                         desc = irq_to_desc(irq);
1133                                         if (desc)
1134                                                 generic_handle_irq_desc(irq, desc);
1135                                 }
1136                         }
1137                 }
1138
1139                 BUG_ON(!irqs_disabled());
1140
1141                 count = __this_cpu_read(xed_nesting_count);
1142                 __this_cpu_write(xed_nesting_count, 0);
1143         } while (count != 1 || vcpu_info->evtchn_upcall_pending);
1144
1145 out:
1146
1147         put_cpu();
1148 }
1149
1150 void xen_evtchn_do_upcall(struct pt_regs *regs)
1151 {
1152         struct pt_regs *old_regs = set_irq_regs(regs);
1153
1154         exit_idle();
1155         irq_enter();
1156
1157         __xen_evtchn_do_upcall();
1158
1159         irq_exit();
1160         set_irq_regs(old_regs);
1161 }
1162
1163 void xen_hvm_evtchn_do_upcall(void)
1164 {
1165         __xen_evtchn_do_upcall();
1166 }
1167 EXPORT_SYMBOL_GPL(xen_hvm_evtchn_do_upcall);
1168
1169 /* Rebind a new event channel to an existing irq. */
1170 void rebind_evtchn_irq(int evtchn, int irq)
1171 {
1172         struct irq_info *info = info_for_irq(irq);
1173
1174         /* Make sure the irq is masked, since the new event channel
1175            will also be masked. */
1176         disable_irq(irq);
1177
1178         spin_lock(&irq_mapping_update_lock);
1179
1180         /* After resume the irq<->evtchn mappings are all cleared out */
1181         BUG_ON(evtchn_to_irq[evtchn] != -1);
1182         /* Expect irq to have been bound before,
1183            so there should be a proper type */
1184         BUG_ON(info->type == IRQT_UNBOUND);
1185
1186         evtchn_to_irq[evtchn] = irq;
1187         irq_info[irq] = mk_evtchn_info(evtchn);
1188
1189         spin_unlock(&irq_mapping_update_lock);
1190
1191         /* new event channels are always bound to cpu 0 */
1192         irq_set_affinity(irq, cpumask_of(0));
1193
1194         /* Unmask the event channel. */
1195         enable_irq(irq);
1196 }
1197
1198 /* Rebind an evtchn so that it gets delivered to a specific cpu */
1199 static int rebind_irq_to_cpu(unsigned irq, unsigned tcpu)
1200 {
1201         struct evtchn_bind_vcpu bind_vcpu;
1202         int evtchn = evtchn_from_irq(irq);
1203
1204         /* events delivered via platform PCI interrupts are always
1205          * routed to vcpu 0 */
1206         if (!VALID_EVTCHN(evtchn) ||
1207                 (xen_hvm_domain() && !xen_have_vector_callback))
1208                 return -1;
1209
1210         /* Send future instances of this interrupt to other vcpu. */
1211         bind_vcpu.port = evtchn;
1212         bind_vcpu.vcpu = tcpu;
1213
1214         /*
1215          * If this fails, it usually just indicates that we're dealing with a
1216          * virq or IPI channel, which don't actually need to be rebound. Ignore
1217          * it, but don't do the xenlinux-level rebind in that case.
1218          */
1219         if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0)
1220                 bind_evtchn_to_cpu(evtchn, tcpu);
1221
1222         return 0;
1223 }
1224
1225 static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
1226                             bool force)
1227 {
1228         unsigned tcpu = cpumask_first(dest);
1229
1230         return rebind_irq_to_cpu(data->irq, tcpu);
1231 }
1232
1233 int resend_irq_on_evtchn(unsigned int irq)
1234 {
1235         int masked, evtchn = evtchn_from_irq(irq);
1236         struct shared_info *s = HYPERVISOR_shared_info;
1237
1238         if (!VALID_EVTCHN(evtchn))
1239                 return 1;
1240
1241         masked = sync_test_and_set_bit(evtchn, s->evtchn_mask);
1242         sync_set_bit(evtchn, s->evtchn_pending);
1243         if (!masked)
1244                 unmask_evtchn(evtchn);
1245
1246         return 1;
1247 }
1248
1249 static void enable_dynirq(struct irq_data *data)
1250 {
1251         int evtchn = evtchn_from_irq(data->irq);
1252
1253         if (VALID_EVTCHN(evtchn))
1254                 unmask_evtchn(evtchn);
1255 }
1256
1257 static void disable_dynirq(struct irq_data *data)
1258 {
1259         int evtchn = evtchn_from_irq(data->irq);
1260
1261         if (VALID_EVTCHN(evtchn))
1262                 mask_evtchn(evtchn);
1263 }
1264
1265 static void ack_dynirq(struct irq_data *data)
1266 {
1267         int evtchn = evtchn_from_irq(data->irq);
1268
1269         move_masked_irq(data->irq);
1270
1271         if (VALID_EVTCHN(evtchn))
1272                 unmask_evtchn(evtchn);
1273 }
1274
1275 static int retrigger_dynirq(struct irq_data *data)
1276 {
1277         int evtchn = evtchn_from_irq(data->irq);
1278         struct shared_info *sh = HYPERVISOR_shared_info;
1279         int ret = 0;
1280
1281         if (VALID_EVTCHN(evtchn)) {
1282                 int masked;
1283
1284                 masked = sync_test_and_set_bit(evtchn, sh->evtchn_mask);
1285                 sync_set_bit(evtchn, sh->evtchn_pending);
1286                 if (!masked)
1287                         unmask_evtchn(evtchn);
1288                 ret = 1;
1289         }
1290
1291         return ret;
1292 }
1293
1294 static void restore_cpu_pirqs(void)
1295 {
1296         int pirq, rc, irq, gsi;
1297         struct physdev_map_pirq map_irq;
1298
1299         for (pirq = 0; pirq < nr_irqs; pirq++) {
1300                 irq = pirq_to_irq[pirq];
1301                 if (irq == -1)
1302                         continue;
1303
1304                 /* save/restore of PT devices doesn't work, so at this point the
1305                  * only devices present are GSI based emulated devices */
1306                 gsi = gsi_from_irq(irq);
1307                 if (!gsi)
1308                         continue;
1309
1310                 map_irq.domid = DOMID_SELF;
1311                 map_irq.type = MAP_PIRQ_TYPE_GSI;
1312                 map_irq.index = gsi;
1313                 map_irq.pirq = pirq;
1314
1315                 rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
1316                 if (rc) {
1317                         printk(KERN_WARNING "xen map irq failed gsi=%d irq=%d pirq=%d rc=%d\n",
1318                                         gsi, irq, pirq, rc);
1319                         irq_info[irq] = mk_unbound_info();
1320                         pirq_to_irq[pirq] = -1;
1321                         continue;
1322                 }
1323
1324                 printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
1325
1326                 __startup_pirq(irq);
1327         }
1328 }
1329
1330 static void restore_cpu_virqs(unsigned int cpu)
1331 {
1332         struct evtchn_bind_virq bind_virq;
1333         int virq, irq, evtchn;
1334
1335         for (virq = 0; virq < NR_VIRQS; virq++) {
1336                 if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
1337                         continue;
1338
1339                 BUG_ON(virq_from_irq(irq) != virq);
1340
1341                 /* Get a new binding from Xen. */
1342                 bind_virq.virq = virq;
1343                 bind_virq.vcpu = cpu;
1344                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
1345                                                 &bind_virq) != 0)
1346                         BUG();
1347                 evtchn = bind_virq.port;
1348
1349                 /* Record the new mapping. */
1350                 evtchn_to_irq[evtchn] = irq;
1351                 irq_info[irq] = mk_virq_info(evtchn, virq);
1352                 bind_evtchn_to_cpu(evtchn, cpu);
1353         }
1354 }
1355
1356 static void restore_cpu_ipis(unsigned int cpu)
1357 {
1358         struct evtchn_bind_ipi bind_ipi;
1359         int ipi, irq, evtchn;
1360
1361         for (ipi = 0; ipi < XEN_NR_IPIS; ipi++) {
1362                 if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
1363                         continue;
1364
1365                 BUG_ON(ipi_from_irq(irq) != ipi);
1366
1367                 /* Get a new binding from Xen. */
1368                 bind_ipi.vcpu = cpu;
1369                 if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
1370                                                 &bind_ipi) != 0)
1371                         BUG();
1372                 evtchn = bind_ipi.port;
1373
1374                 /* Record the new mapping. */
1375                 evtchn_to_irq[evtchn] = irq;
1376                 irq_info[irq] = mk_ipi_info(evtchn, ipi);
1377                 bind_evtchn_to_cpu(evtchn, cpu);
1378         }
1379 }
1380
1381 /* Clear an irq's pending state, in preparation for polling on it */
1382 void xen_clear_irq_pending(int irq)
1383 {
1384         int evtchn = evtchn_from_irq(irq);
1385
1386         if (VALID_EVTCHN(evtchn))
1387                 clear_evtchn(evtchn);
1388 }
1389 EXPORT_SYMBOL(xen_clear_irq_pending);
1390 void xen_set_irq_pending(int irq)
1391 {
1392         int evtchn = evtchn_from_irq(irq);
1393
1394         if (VALID_EVTCHN(evtchn))
1395                 set_evtchn(evtchn);
1396 }
1397
1398 bool xen_test_irq_pending(int irq)
1399 {
1400         int evtchn = evtchn_from_irq(irq);
1401         bool ret = false;
1402
1403         if (VALID_EVTCHN(evtchn))
1404                 ret = test_evtchn(evtchn);
1405
1406         return ret;
1407 }
1408
1409 /* Poll waiting for an irq to become pending with timeout.  In the usual case,
1410  * the irq will be disabled so it won't deliver an interrupt. */
1411 void xen_poll_irq_timeout(int irq, u64 timeout)
1412 {
1413         evtchn_port_t evtchn = evtchn_from_irq(irq);
1414
1415         if (VALID_EVTCHN(evtchn)) {
1416                 struct sched_poll poll;
1417
1418                 poll.nr_ports = 1;
1419                 poll.timeout = timeout;
1420                 set_xen_guest_handle(poll.ports, &evtchn);
1421
1422                 if (HYPERVISOR_sched_op(SCHEDOP_poll, &poll) != 0)
1423                         BUG();
1424         }
1425 }
1426 EXPORT_SYMBOL(xen_poll_irq_timeout);
1427 /* Poll waiting for an irq to become pending.  In the usual case, the
1428  * irq will be disabled so it won't deliver an interrupt. */
1429 void xen_poll_irq(int irq)
1430 {
1431         xen_poll_irq_timeout(irq, 0 /* no timeout */);
1432 }
1433
1434 void xen_irq_resume(void)
1435 {
1436         unsigned int cpu, irq, evtchn;
1437
1438         init_evtchn_cpu_bindings();
1439
1440         /* New event-channel space is not 'live' yet. */
1441         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1442                 mask_evtchn(evtchn);
1443
1444         /* No IRQ <-> event-channel mappings. */
1445         for (irq = 0; irq < nr_irqs; irq++)
1446                 irq_info[irq].evtchn = 0; /* zap event-channel binding */
1447
1448         for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
1449                 evtchn_to_irq[evtchn] = -1;
1450
1451         for_each_possible_cpu(cpu) {
1452                 restore_cpu_virqs(cpu);
1453                 restore_cpu_ipis(cpu);
1454         }
1455
1456         restore_cpu_pirqs();
1457 }
1458
1459 static struct irq_chip xen_dynamic_chip __read_mostly = {
1460         .name                   = "xen-dyn",
1461
1462         .irq_disable            = disable_dynirq,
1463         .irq_mask               = disable_dynirq,
1464         .irq_unmask             = enable_dynirq,
1465
1466         .irq_eoi                = ack_dynirq,
1467         .irq_set_affinity       = set_affinity_irq,
1468         .irq_retrigger          = retrigger_dynirq,
1469 };
1470
1471 static struct irq_chip xen_pirq_chip __read_mostly = {
1472         .name                   = "xen-pirq",
1473
1474         .irq_startup            = startup_pirq,
1475         .irq_shutdown           = shutdown_pirq,
1476
1477         .irq_enable             = enable_pirq,
1478         .irq_unmask             = enable_pirq,
1479
1480         .irq_disable            = disable_pirq,
1481         .irq_mask               = disable_pirq,
1482
1483         .irq_ack                = ack_pirq,
1484
1485         .irq_set_affinity       = set_affinity_irq,
1486
1487         .irq_retrigger          = retrigger_dynirq,
1488 };
1489
1490 static struct irq_chip xen_percpu_chip __read_mostly = {
1491         .name                   = "xen-percpu",
1492
1493         .irq_disable            = disable_dynirq,
1494         .irq_mask               = disable_dynirq,
1495         .irq_unmask             = enable_dynirq,
1496
1497         .irq_ack                = ack_dynirq,
1498 };
1499
1500 int xen_set_callback_via(uint64_t via)
1501 {
1502         struct xen_hvm_param a;
1503         a.domid = DOMID_SELF;
1504         a.index = HVM_PARAM_CALLBACK_IRQ;
1505         a.value = via;
1506         return HYPERVISOR_hvm_op(HVMOP_set_param, &a);
1507 }
1508 EXPORT_SYMBOL_GPL(xen_set_callback_via);
1509
1510 #ifdef CONFIG_XEN_PVHVM
1511 /* Vector callbacks are better than PCI interrupts to receive event
1512  * channel notifications because we can receive vector callbacks on any
1513  * vcpu and we don't need PCI support or APIC interactions. */
1514 void xen_callback_vector(void)
1515 {
1516         int rc;
1517         uint64_t callback_via;
1518         if (xen_have_vector_callback) {
1519                 callback_via = HVM_CALLBACK_VECTOR(XEN_HVM_EVTCHN_CALLBACK);
1520                 rc = xen_set_callback_via(callback_via);
1521                 if (rc) {
1522                         printk(KERN_ERR "Request for Xen HVM callback vector"
1523                                         " failed.\n");
1524                         xen_have_vector_callback = 0;
1525                         return;
1526                 }
1527                 printk(KERN_INFO "Xen HVM callback vector for event delivery is "
1528                                 "enabled\n");
1529                 /* in the restore case the vector has already been allocated */
1530                 if (!test_bit(XEN_HVM_EVTCHN_CALLBACK, used_vectors))
1531                         alloc_intr_gate(XEN_HVM_EVTCHN_CALLBACK, xen_hvm_callback_vector);
1532         }
1533 }
1534 #else
1535 void xen_callback_vector(void) {}
1536 #endif
1537
1538 void __init xen_init_IRQ(void)
1539 {
1540         int i;
1541
1542         cpu_evtchn_mask_p = kcalloc(nr_cpu_ids, sizeof(struct cpu_evtchn_s),
1543                                     GFP_KERNEL);
1544         irq_info = kcalloc(nr_irqs, sizeof(*irq_info), GFP_KERNEL);
1545
1546         /* We are using nr_irqs as the maximum number of pirq available but
1547          * that number is actually chosen by Xen and we don't know exactly
1548          * what it is. Be careful choosing high pirq numbers. */
1549         pirq_to_irq = kcalloc(nr_irqs, sizeof(*pirq_to_irq), GFP_KERNEL);
1550         for (i = 0; i < nr_irqs; i++)
1551                 pirq_to_irq[i] = -1;
1552
1553         evtchn_to_irq = kcalloc(NR_EVENT_CHANNELS, sizeof(*evtchn_to_irq),
1554                                     GFP_KERNEL);
1555         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1556                 evtchn_to_irq[i] = -1;
1557
1558         init_evtchn_cpu_bindings();
1559
1560         /* No event channels are 'live' right now. */
1561         for (i = 0; i < NR_EVENT_CHANNELS; i++)
1562                 mask_evtchn(i);
1563
1564         if (xen_hvm_domain()) {
1565                 xen_callback_vector();
1566                 native_init_IRQ();
1567                 /* pci_xen_hvm_init must be called after native_init_IRQ so that
1568                  * __acpi_register_gsi can point at the right function */
1569                 pci_xen_hvm_init();
1570         } else {
1571                 irq_ctx_init(smp_processor_id());
1572                 if (xen_initial_domain())
1573                         xen_setup_pirqs();
1574         }
1575 }